Shares of Arm Holdings moved slightly higher in premarket trading after Bank of America outlined a more optimistic medium- to long-term CPU market scenario in a note to investors.
Analyst Vivek Arya said the characteristics of AI inference - which he described as "control-heavy" - increase the dependence on CPUs, and that CPUs will continue to represent roughly 4-5% of the overall AI data-center total addressable market (TAM). BofA places that total AI data-center TAM at about $1.4 trillion.
On the server-CPU side, BofA projects notable expansion. The bank estimates server CPU TAM could grow from about $27 billion in calendar year 2025 to roughly $60 billion by CY30E, a compound annual growth rate of approximately 17%. BofA expects AI servers to account for nearly 70% of that server CPU market by CY30E.
Against that backdrop, BofA lifted its view of Arm's potential CPU share gains. The bank now sees the possibility of Arm reaching a roughly 20-25%+ share of server CPUs by CY30E, up from a prior projection of 15-20%. BofA framed this as a response to AI's shift "away from just training and toward inference," a phase that the bank says relies more heavily on control-oriented CPU performance.
The note also pointed to concrete market momentum: record share gains in the fourth quarter of 2025, with Arm achieving a record 11.9% server value share, according to BofA's commentary.
BofA further suggested that Arm's planned entry into merchant CPUs could materially enlarge its addressable market. The bank estimated that Arm's serviceable available market (SAM) could expand by a factor of 30 if the company were to deploy full chiplet-based solutions.
At the same time, BofA warned of an offsetting factor: the potential loss of current SoftBank licensing revenue, which it quantified at about $800 million per year. The bank said that this possible reduction in licensing sales could partly counterbalance medium-term gains from the CPU opportunity.
Balancing these drivers and headwinds, BofA kept its Neutral rating on Arm but raised the price target to $140, citing the improving long-term CPU share opportunity as the rationale for the higher target.
Sector implications: The analysis touches primarily on semiconductors and data-center hardware markets, with secondary effects for cloud infrastructure and AI server suppliers. Growing CPU demand for inference workloads would influence vendor road maps and capital allocation decisions across these sectors.